fs/btrfs/transaction.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/sched.h>
  21 #include <linux/writeback.h>
  22 #include <linux/pagemap.h>
  23 #include "ctree.h"
  24 #include "disk-io.h"
  25 #include "transaction.h"
  26
  27 static int total_trans = 0;
  28 extern struct kmem_cache *btrfs_trans_handle_cachep;
  29 extern struct kmem_cache *btrfs_transaction_cachep;
  30
  31 static struct workqueue_struct *trans_wq;
  32
  33 #define BTRFS_ROOT_TRANS_TAG 0
  34 #define BTRFS_ROOT_DEFRAG_TAG 1
  35
  36 static noinline void put_transaction(struct btrfs_transaction *transaction)
  37 {
  38         WARN_ON(transaction->use_count == 0);
  39         transaction->use_count--;
  40         if (transaction->use_count == 0) {
  41                 WARN_ON(total_trans == 0);
  42                 total_trans--;
  43                 list_del_init(&transaction->list);
  44                 memset(transaction, 0, sizeof(*transaction));
  45                 kmem_cache_free(btrfs_transaction_cachep, transaction);
  46         }
  47 }
  48
  49 static noinline int join_transaction(struct btrfs_root *root)
  50 {
  51         struct btrfs_transaction *cur_trans;
  52         cur_trans = root->fs_info->running_transaction;
  53         if (!cur_trans) {
  54                 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
  55                                              GFP_NOFS);
  56                 total_trans++;
  57                 BUG_ON(!cur_trans);
  58                 root->fs_info->generation++;
  59                 root->fs_info->last_alloc = 0;
  60                 root->fs_info->last_data_alloc = 0;
  61                 cur_trans->num_writers = 1;
  62                 cur_trans->num_joined = 0;
  63                 cur_trans->transid = root->fs_info->generation;
  64                 init_waitqueue_head(&cur_trans->writer_wait);
  65                 init_waitqueue_head(&cur_trans->commit_wait);
  66                 cur_trans->in_commit = 0;
  67                 cur_trans->use_count = 1;
  68                 cur_trans->commit_done = 0;
  69                 cur_trans->start_time = get_seconds();
  70                 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
  71                 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
  72                 btrfs_ordered_inode_tree_init(&cur_trans->ordered_inode_tree);
  73                 extent_io_tree_init(&cur_trans->dirty_pages,
  74                                      root->fs_info->btree_inode->i_mapping,
  75                                      GFP_NOFS);
  76                 spin_lock(&root->fs_info->new_trans_lock);
  77                 root->fs_info->running_transaction = cur_trans;
  78                 spin_unlock(&root->fs_info->new_trans_lock);
  79         } else {
  80                 cur_trans->num_writers++;
  81                 cur_trans->num_joined++;
  82         }
  83
  84         return 0;
  85 }
  86
  87 static noinline int record_root_in_trans(struct btrfs_root *root)
  88 {
  89         u64 running_trans_id = root->fs_info->running_transaction->transid;
  90         if (root->ref_cows && root->last_trans < running_trans_id) {
  91                 WARN_ON(root == root->fs_info->extent_root);
  92                 if (root->root_item.refs != 0) {
  93                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
  94                                    (unsigned long)root->root_key.objectid,
  95                                    BTRFS_ROOT_TRANS_TAG);
  96                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
  97                                    (unsigned long)root->root_key.objectid,
  98                                    BTRFS_ROOT_DEFRAG_TAG);
  99                         root->commit_root = root->node;
 100                         extent_buffer_get(root->node);
 101                 } else {
 102                         WARN_ON(1);
 103                 }
 104                 root->last_trans = running_trans_id;
 105         }
 106         return 0;
 107 }
 108
 109 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
 110                                                    int num_blocks)
 111 {
 112         struct btrfs_trans_handle *h =
 113                 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
 114         int ret;
 115
 116         mutex_lock(&root->fs_info->trans_mutex);
 117         ret = join_transaction(root);
 118         BUG_ON(ret);
 119
 120         record_root_in_trans(root);
 121         h->transid = root->fs_info->running_transaction->transid;
 122         h->transaction = root->fs_info->running_transaction;
 123         h->blocks_reserved = num_blocks;
 124         h->blocks_used = 0;
 125         h->block_group = NULL;
 126         h->alloc_exclude_nr = 0;
 127         h->alloc_exclude_start = 0;
 128         root->fs_info->running_transaction->use_count++;
 129         mutex_unlock(&root->fs_info->trans_mutex);
 130         return h;
 131 }
 132
 133 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 134                           struct btrfs_root *root)
 135 {
 136         struct btrfs_transaction *cur_trans;
 137
 138         mutex_lock(&root->fs_info->trans_mutex);
 139         cur_trans = root->fs_info->running_transaction;
 140         WARN_ON(cur_trans != trans->transaction);
 141         WARN_ON(cur_trans->num_writers < 1);
 142         cur_trans->num_writers--;
 143         if (waitqueue_active(&cur_trans->writer_wait))
 144                 wake_up(&cur_trans->writer_wait);
 145         put_transaction(cur_trans);
 146         mutex_unlock(&root->fs_info->trans_mutex);
 147         memset(trans, 0, sizeof(*trans));
 148         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 149         return 0;
 150 }
 151
 152
 153 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 154                                      struct btrfs_root *root)
 155 {
 156         int ret;
 157         int err;
 158         int werr = 0;
 159         struct extent_io_tree *dirty_pages;
 160         struct page *page;
 161         struct inode *btree_inode = root->fs_info->btree_inode;
 162         u64 start;
 163         u64 end;
 164         unsigned long index;
 165
 166         if (!trans || !trans->transaction) {
 167                 return filemap_write_and_wait(btree_inode->i_mapping);
 168         }
 169         dirty_pages = &trans->transaction->dirty_pages;
 170         while(1) {
 171                 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
 172                                             EXTENT_DIRTY);
 173                 if (ret)
 174                         break;
 175                 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
 176                 while(start <= end) {
 177                         index = start >> PAGE_CACHE_SHIFT;
 178                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
 179                         page = find_lock_page(btree_inode->i_mapping, index);
 180                         if (!page)
 181                                 continue;
 182                         if (PageWriteback(page)) {
 183                                 if (PageDirty(page))
 184                                         wait_on_page_writeback(page);
 185                                 else {
 186                                         unlock_page(page);
 187                                         page_cache_release(page);
 188                                         continue;
 189                                 }
 190                         }
 191                         err = write_one_page(page, 0);
 192                         if (err)
 193                                 werr = err;
 194                         page_cache_release(page);
 195                 }
 196         }
 197         err = filemap_fdatawait(btree_inode->i_mapping);
 198         if (err)
 199                 werr = err;
 200         return werr;
 201 }
 202
 203 static int update_cowonly_root(struct btrfs_trans_handle *trans,
 204                                struct btrfs_root *root)
 205 {
 206         int ret;
 207         u64 old_root_bytenr;
 208         struct btrfs_root *tree_root = root->fs_info->tree_root;
 209
 210         btrfs_write_dirty_block_groups(trans, root);
 211         while(1) {
 212                 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
 213                 if (old_root_bytenr == root->node->start)
 214                         break;
 215                 btrfs_set_root_bytenr(&root->root_item,
 216                                        root->node->start);
 217                 btrfs_set_root_level(&root->root_item,
 218                                      btrfs_header_level(root->node));
 219                 ret = btrfs_update_root(trans, tree_root,
 220                                         &root->root_key,
 221                                         &root->root_item);
 222                 BUG_ON(ret);
 223                 btrfs_write_dirty_block_groups(trans, root);
 224         }
 225         return 0;
 226 }
 227
 228 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 229                             struct btrfs_root *root)
 230 {
 231         struct btrfs_fs_info *fs_info = root->fs_info;
 232         struct list_head *next;
 233
 234         while(!list_empty(&fs_info->dirty_cowonly_roots)) {
 235                 next = fs_info->dirty_cowonly_roots.next;
 236                 list_del_init(next);
 237                 root = list_entry(next, struct btrfs_root, dirty_list);
 238                 update_cowonly_root(trans, root);
 239         }
 240         return 0;
 241 }
 242
 243 static noinline int wait_for_commit(struct btrfs_root *root,
 244                                     struct btrfs_transaction *commit)
 245 {
 246         DEFINE_WAIT(wait);
 247         mutex_lock(&root->fs_info->trans_mutex);
 248         while(!commit->commit_done) {
 249                 prepare_to_wait(&commit->commit_wait, &wait,
 250                                 TASK_UNINTERRUPTIBLE);
 251                 if (commit->commit_done)
 252                         break;
 253                 mutex_unlock(&root->fs_info->trans_mutex);
 254                 schedule();
 255                 mutex_lock(&root->fs_info->trans_mutex);
 256         }
 257         mutex_unlock(&root->fs_info->trans_mutex);
 258         finish_wait(&commit->commit_wait, &wait);
 259         return 0;
 260 }
 261
 262 struct dirty_root {
 263         struct list_head list;
 264         struct btrfs_root *root;
 265         struct btrfs_root *latest_root;
 266 };
 267
 268 int btrfs_add_dead_root(struct btrfs_root *root,
 269                         struct btrfs_root *latest,
 270                         struct list_head *dead_list)
 271 {
 272         struct dirty_root *dirty;
 273
 274         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 275         if (!dirty)
 276                 return -ENOMEM;
 277         dirty->root = root;
 278         dirty->latest_root = latest;
 279         list_add(&dirty->list, dead_list);
 280         return 0;
 281 }
 282
 283 static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
 284                                     struct radix_tree_root *radix,
 285                                     struct list_head *list)
 286 {
 287         struct dirty_root *dirty;
 288         struct btrfs_root *gang[8];
 289         struct btrfs_root *root;
 290         int i;
 291         int ret;
 292         int err = 0;
 293         u32 refs;
 294
 295         while(1) {
 296                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
 297                                                  ARRAY_SIZE(gang),
 298                                                  BTRFS_ROOT_TRANS_TAG);
 299                 if (ret == 0)
 300                         break;
 301                 for (i = 0; i < ret; i++) {
 302                         root = gang[i];
 303                         radix_tree_tag_clear(radix,
 304                                      (unsigned long)root->root_key.objectid,
 305                                      BTRFS_ROOT_TRANS_TAG);
 306                         if (root->commit_root == root->node) {
 307                                 WARN_ON(root->node->start !=
 308                                         btrfs_root_bytenr(&root->root_item));
 309                                 free_extent_buffer(root->commit_root);
 310                                 root->commit_root = NULL;
 311
 312                                 /* make sure to update the root on disk
 313                                  * so we get any updates to the block used
 314                                  * counts
 315                                  */
 316                                 err = btrfs_update_root(trans,
 317                                                 root->fs_info->tree_root,
 318                                                 &root->root_key,
 319                                                 &root->root_item);
 320                                 continue;
 321                         }
 322                         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
 323                         BUG_ON(!dirty);
 324                         dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
 325                         BUG_ON(!dirty->root);
 326
 327                         memset(&root->root_item.drop_progress, 0,
 328                                sizeof(struct btrfs_disk_key));
 329                         root->root_item.drop_level = 0;
 330
 331                         memcpy(dirty->root, root, sizeof(*root));
 332                         dirty->root->node = root->commit_root;
 333                         dirty->latest_root = root;
 334                         root->commit_root = NULL;
 335
 336                         root->root_key.offset = root->fs_info->generation;
 337                         btrfs_set_root_bytenr(&root->root_item,
 338                                               root->node->start);
 339                         btrfs_set_root_level(&root->root_item,
 340                                              btrfs_header_level(root->node));
 341                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
 342                                                 &root->root_key,
 343                                                 &root->root_item);
 344                         if (err)
 345                                 break;
 346
 347                         refs = btrfs_root_refs(&dirty->root->root_item);
 348                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
 349                         err = btrfs_update_root(trans, root->fs_info->tree_root,
 350                                                 &dirty->root->root_key,
 351                                                 &dirty->root->root_item);
 352
 353                         BUG_ON(err);
 354                         if (refs == 1) {
 355                                 list_add(&dirty->list, list);
 356                         } else {
 357                                 WARN_ON(1);
 358                                 kfree(dirty->root);
 359                                 kfree(dirty);
 360                         }
 361                 }
 362         }
 363         return err;
 364 }
 365
 366 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 367 {
 368         struct btrfs_fs_info *info = root->fs_info;
 369         int ret;
 370         struct btrfs_trans_handle *trans;
 371         unsigned long nr;
 372
 373         if (root->defrag_running)
 374                 return 0;
 375         trans = btrfs_start_transaction(root, 1);
 376         while (1) {
 377                 root->defrag_running = 1;
 378                 ret = btrfs_defrag_leaves(trans, root, cacheonly);
 379                 nr = trans->blocks_used;
 380                 btrfs_end_transaction(trans, root);
 381                 mutex_unlock(&info->fs_mutex);
 382                 btrfs_btree_balance_dirty(info->tree_root, nr);
 383                 cond_resched();
 384
 385                 mutex_lock(&info->fs_mutex);
 386                 trans = btrfs_start_transaction(root, 1);
 387                 if (ret != -EAGAIN)
 388                         break;
 389         }
 390         root->defrag_running = 0;
 391         radix_tree_tag_clear(&info->fs_roots_radix,
 392                      (unsigned long)root->root_key.objectid,
 393                      BTRFS_ROOT_DEFRAG_TAG);
 394         btrfs_end_transaction(trans, root);
 395         return 0;
 396 }
 397
 398 int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
 399 {
 400         struct btrfs_root *gang[1];
 401         struct btrfs_root *root;
 402         int i;
 403         int ret;
 404         int err = 0;
 405         u64 last = 0;
 406
 407         while(1) {
 408                 ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
 409                                                  (void **)gang, last,
 410                                                  ARRAY_SIZE(gang),
 411                                                  BTRFS_ROOT_DEFRAG_TAG);
 412                 if (ret == 0)
 413                         break;
 414                 for (i = 0; i < ret; i++) {
 415                         root = gang[i];
 416                         last = root->root_key.objectid + 1;
 417                         btrfs_defrag_root(root, 1);
 418                 }
 419         }
 420         btrfs_defrag_root(info->extent_root, 1);
 421         return err;
 422 }
 423
 424 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
 425                                      struct list_head *list)
 426 {
 427         struct dirty_root *dirty;
 428         struct btrfs_trans_handle *trans;
 429         unsigned long nr;
 430         u64 num_bytes;
 431         u64 bytes_used;
 432         int ret = 0;
 433         int err;
 434
 435         while(!list_empty(list)) {
 436                 struct btrfs_root *root;
 437
 438                 mutex_lock(&tree_root->fs_info->fs_mutex);
 439                 dirty = list_entry(list->next, struct dirty_root, list);
 440                 list_del_init(&dirty->list);
 441
 442                 num_bytes = btrfs_root_used(&dirty->root->root_item);
 443                 root = dirty->latest_root;
 444                 root->fs_info->throttles++;
 445
 446                 while(1) {
 447                         trans = btrfs_start_transaction(tree_root, 1);
 448                         ret = btrfs_drop_snapshot(trans, dirty->root);
 449                         if (ret != -EAGAIN) {
 450                                 break;
 451                         }
 452
 453                         err = btrfs_update_root(trans,
 454                                         tree_root,
 455                                         &dirty->root->root_key,
 456                                         &dirty->root->root_item);
 457                         if (err)
 458                                 ret = err;
 459                         nr = trans->blocks_used;
 460                         ret = btrfs_end_transaction(trans, tree_root);
 461                         BUG_ON(ret);
 462                         mutex_unlock(&tree_root->fs_info->fs_mutex);
 463                         btrfs_btree_balance_dirty(tree_root, nr);
 464                         cond_resched();
 465                         mutex_lock(&tree_root->fs_info->fs_mutex);
 466                 }
 467                 BUG_ON(ret);
 468                 root->fs_info->throttles--;
 469
 470                 num_bytes -= btrfs_root_used(&dirty->root->root_item);
 471                 bytes_used = btrfs_root_used(&root->root_item);
 472                 if (num_bytes) {
 473                         record_root_in_trans(root);
 474                         btrfs_set_root_used(&root->root_item,
 475                                             bytes_used - num_bytes);
 476                 }
 477                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
 478                 if (ret) {
 479                         BUG();
 480                         break;
 481                 }
 482                 nr = trans->blocks_used;
 483                 ret = btrfs_end_transaction(trans, tree_root);
 484                 BUG_ON(ret);
 485
 486                 free_extent_buffer(dirty->root->node);
 487                 kfree(dirty->root);
 488                 kfree(dirty);
 489                 mutex_unlock(&tree_root->fs_info->fs_mutex);
 490
 491                 btrfs_btree_balance_dirty(tree_root, nr);
 492                 cond_resched();
 493         }
 494         return ret;
 495 }
 496
 497 int btrfs_write_ordered_inodes(struct btrfs_trans_handle *trans,
 498                                 struct btrfs_root *root)
 499 {
 500         struct btrfs_transaction *cur_trans = trans->transaction;
 501         struct inode *inode;
 502         u64 root_objectid = 0;
 503         u64 objectid = 0;
 504         int ret;
 505
 506         root->fs_info->throttles++;
 507         while(1) {
 508                 ret = btrfs_find_first_ordered_inode(
 509                                 &cur_trans->ordered_inode_tree,
 510                                 &root_objectid, &objectid, &inode);
 511                 if (!ret)
 512                         break;
 513
 514                 mutex_unlock(&root->fs_info->trans_mutex);
 515                 mutex_unlock(&root->fs_info->fs_mutex);
 516
 517                 if (S_ISREG(inode->i_mode)) {
 518                         atomic_inc(&BTRFS_I(inode)->ordered_writeback);
 519                         filemap_fdatawrite(inode->i_mapping);
 520                         atomic_dec(&BTRFS_I(inode)->ordered_writeback);
 521                 }
 522                 iput(inode);
 523
 524                 mutex_lock(&root->fs_info->fs_mutex);
 525                 mutex_lock(&root->fs_info->trans_mutex);
 526         }
 527         while(1) {
 528                 root_objectid = 0;
 529                 objectid = 0;
 530                 ret = btrfs_find_del_first_ordered_inode(
 531                                 &cur_trans->ordered_inode_tree,
 532                                 &root_objectid, &objectid, &inode);
 533                 if (!ret)
 534                         break;
 535                 mutex_unlock(&root->fs_info->trans_mutex);
 536                 mutex_unlock(&root->fs_info->fs_mutex);
 537
 538                 if (S_ISREG(inode->i_mode)) {
 539                         atomic_inc(&BTRFS_I(inode)->ordered_writeback);
 540                         filemap_write_and_wait(inode->i_mapping);
 541                         atomic_dec(&BTRFS_I(inode)->ordered_writeback);
 542                 }
 543                 atomic_dec(&inode->i_count);
 544                 iput(inode);
 545
 546                 mutex_lock(&root->fs_info->fs_mutex);
 547                 mutex_lock(&root->fs_info->trans_mutex);
 548         }
 549         root->fs_info->throttles--;
 550         return 0;
 551 }
 552
 553 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 554                                    struct btrfs_fs_info *fs_info,
 555                                    struct btrfs_pending_snapshot *pending)
 556 {
 557         struct btrfs_key key;
 558         struct btrfs_root_item *new_root_item;
 559         struct btrfs_root *tree_root = fs_info->tree_root;
 560         struct btrfs_root *root = pending->root;
 561         struct extent_buffer *tmp;
 562         int ret;
 563         u64 objectid;
 564
 565         new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
 566         if (!new_root_item) {
 567                 ret = -ENOMEM;
 568                 goto fail;
 569         }
 570         ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
 571         if (ret)
 572                 goto fail;
 573
 574         memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 575
 576         key.objectid = objectid;
 577         key.offset = 1;
 578         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 579
 580         extent_buffer_get(root->node);
 581         btrfs_cow_block(trans, root, root->node, NULL, 0, &tmp);
 582         free_extent_buffer(tmp);
 583
 584         btrfs_copy_root(trans, root, root->node, &tmp, objectid);
 585
 586         btrfs_set_root_bytenr(new_root_item, tmp->start);
 587         btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
 588         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
 589                                 new_root_item);
 590         free_extent_buffer(tmp);
 591         if (ret)
 592                 goto fail;
 593
 594         /*
 595          * insert the directory item
 596          */
 597         key.offset = (u64)-1;
 598         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
 599                                     pending->name, strlen(pending->name),
 600                                     root->fs_info->sb->s_root->d_inode->i_ino,
 601                                     &key, BTRFS_FT_DIR);
 602
 603         if (ret)
 604                 goto fail;
 605
 606         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
 607                              pending->name, strlen(pending->name), objectid,
 608                              root->fs_info->sb->s_root->d_inode->i_ino);
 609 fail:
 610         kfree(new_root_item);
 611         return ret;
 612 }
 613
 614 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
 615                                              struct btrfs_fs_info *fs_info)
 616 {
 617         struct btrfs_pending_snapshot *pending;
 618         struct list_head *head = &trans->transaction->pending_snapshots;
 619         int ret;
 620
 621         while(!list_empty(head)) {
 622                 pending = list_entry(head->next,
 623                                      struct btrfs_pending_snapshot, list);
 624                 ret = create_pending_snapshot(trans, fs_info, pending);
 625                 BUG_ON(ret);
 626                 list_del(&pending->list);
 627                 kfree(pending->name);
 628                 kfree(pending);
 629         }
 630         return 0;
 631 }
 632
 633 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 634                              struct btrfs_root *root)
 635 {
 636         unsigned long joined = 0;
 637         unsigned long timeout = 1;
 638         struct btrfs_transaction *cur_trans;
 639         struct btrfs_transaction *prev_trans = NULL;
 640         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
 641         struct list_head dirty_fs_roots;
 642         struct extent_io_tree *pinned_copy;
 643         DEFINE_WAIT(wait);
 644         int ret;
 645
 646         INIT_LIST_HEAD(&dirty_fs_roots);
 647
 648         mutex_lock(&root->fs_info->trans_mutex);
 649         if (trans->transaction->in_commit) {
 650                 cur_trans = trans->transaction;
 651                 trans->transaction->use_count++;
 652                 mutex_unlock(&root->fs_info->trans_mutex);
 653                 btrfs_end_transaction(trans, root);
 654
 655                 mutex_unlock(&root->fs_info->fs_mutex);
 656                 ret = wait_for_commit(root, cur_trans);
 657                 BUG_ON(ret);
 658
 659                 mutex_lock(&root->fs_info->trans_mutex);
 660                 put_transaction(cur_trans);
 661                 mutex_unlock(&root->fs_info->trans_mutex);
 662
 663                 mutex_lock(&root->fs_info->fs_mutex);
 664                 return 0;
 665         }
 666
 667         pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
 668         if (!pinned_copy)
 669                 return -ENOMEM;
 670
 671         extent_io_tree_init(pinned_copy,
 672                              root->fs_info->btree_inode->i_mapping, GFP_NOFS);
 673
 674         trans->transaction->in_commit = 1;
 675         cur_trans = trans->transaction;
 676         if (cur_trans->list.prev != &root->fs_info->trans_list) {
 677                 prev_trans = list_entry(cur_trans->list.prev,
 678                                         struct btrfs_transaction, list);
 679                 if (!prev_trans->commit_done) {
 680                         prev_trans->use_count++;
 681                         mutex_unlock(&root->fs_info->fs_mutex);
 682                         mutex_unlock(&root->fs_info->trans_mutex);
 683
 684                         wait_for_commit(root, prev_trans);
 685
 686                         mutex_lock(&root->fs_info->fs_mutex);
 687                         mutex_lock(&root->fs_info->trans_mutex);
 688                         put_transaction(prev_trans);
 689                 }
 690         }
 691
 692         do {
 693                 joined = cur_trans->num_joined;
 694                 WARN_ON(cur_trans != trans->transaction);
 695                 prepare_to_wait(&cur_trans->writer_wait, &wait,
 696                                 TASK_UNINTERRUPTIBLE);
 697
 698                 if (cur_trans->num_writers > 1)
 699                         timeout = MAX_SCHEDULE_TIMEOUT;
 700                 else
 701                         timeout = 1;
 702
 703                 mutex_unlock(&root->fs_info->fs_mutex);
 704                 mutex_unlock(&root->fs_info->trans_mutex);
 705
 706                 schedule_timeout(timeout);
 707
 708                 mutex_lock(&root->fs_info->fs_mutex);
 709                 mutex_lock(&root->fs_info->trans_mutex);
 710                 finish_wait(&cur_trans->writer_wait, &wait);
 711                 ret = btrfs_write_ordered_inodes(trans, root);
 712
 713         } while (cur_trans->num_writers > 1 ||
 714                  (cur_trans->num_joined != joined));
 715
 716         ret = create_pending_snapshots(trans, root->fs_info);
 717         BUG_ON(ret);
 718
 719         WARN_ON(cur_trans != trans->transaction);
 720
 721         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
 722                               &dirty_fs_roots);
 723         BUG_ON(ret);
 724
 725         ret = btrfs_commit_tree_roots(trans, root);
 726         BUG_ON(ret);
 727
 728         cur_trans = root->fs_info->running_transaction;
 729         spin_lock(&root->fs_info->new_trans_lock);
 730         root->fs_info->running_transaction = NULL;
 731         spin_unlock(&root->fs_info->new_trans_lock);
 732         btrfs_set_super_generation(&root->fs_info->super_copy,
 733                                    cur_trans->transid);
 734         btrfs_set_super_root(&root->fs_info->super_copy,
 735                              root->fs_info->tree_root->node->start);
 736         btrfs_set_super_root_level(&root->fs_info->super_copy,
 737                            btrfs_header_level(root->fs_info->tree_root->node));
 738
 739         btrfs_set_super_chunk_root(&root->fs_info->super_copy,
 740                                    chunk_root->node->start);
 741         btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
 742                                          btrfs_header_level(chunk_root->node));
 743         memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
 744                sizeof(root->fs_info->super_copy));
 745
 746         btrfs_copy_pinned(root, pinned_copy);
 747
 748         mutex_unlock(&root->fs_info->trans_mutex);
 749         mutex_unlock(&root->fs_info->fs_mutex);
 750         ret = btrfs_write_and_wait_transaction(trans, root);
 751         BUG_ON(ret);
 752         write_ctree_super(trans, root);
 753
 754         mutex_lock(&root->fs_info->fs_mutex);
 755         btrfs_finish_extent_commit(trans, root, pinned_copy);
 756         mutex_lock(&root->fs_info->trans_mutex);
 757
 758         kfree(pinned_copy);
 759
 760         cur_trans->commit_done = 1;
 761         root->fs_info->last_trans_committed = cur_trans->transid;
 762         wake_up(&cur_trans->commit_wait);
 763         put_transaction(cur_trans);
 764         put_transaction(cur_trans);
 765
 766         if (root->fs_info->closing)
 767                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
 768         else
 769                 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
 770
 771         mutex_unlock(&root->fs_info->trans_mutex);
 772         kmem_cache_free(btrfs_trans_handle_cachep, trans);
 773
 774         if (root->fs_info->closing) {
 775                 mutex_unlock(&root->fs_info->fs_mutex);
 776                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
 777                 mutex_lock(&root->fs_info->fs_mutex);
 778         }
 779         return ret;
 780 }
 781
 782 int btrfs_clean_old_snapshots(struct btrfs_root *root)
 783 {
 784         struct list_head dirty_roots;
 785         INIT_LIST_HEAD(&dirty_roots);
 786
 787         mutex_lock(&root->fs_info->trans_mutex);
 788         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
 789         mutex_unlock(&root->fs_info->trans_mutex);
 790
 791         if (!list_empty(&dirty_roots)) {
 792                 drop_dirty_roots(root, &dirty_roots);
 793         }
 794         return 0;
 795 }
 796 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 797 void btrfs_transaction_cleaner(void *p)
 798 #else
 799 void btrfs_transaction_cleaner(struct work_struct *work)
 800 #endif
 801 {
 802 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 803         struct btrfs_fs_info *fs_info = p;
 804 #else
 805         struct btrfs_fs_info *fs_info = container_of(work,
 806                                                      struct btrfs_fs_info,
 807                                                      trans_work.work);
 808
 809 #endif
 810         struct btrfs_root *root = fs_info->tree_root;
 811         struct btrfs_transaction *cur;
 812         struct btrfs_trans_handle *trans;
 813         unsigned long now;
 814         unsigned long delay = HZ * 30;
 815         int ret;
 816
 817         mutex_lock(&root->fs_info->fs_mutex);
 818         if (root->fs_info->closing)
 819                 goto out;
 820
 821         mutex_lock(&root->fs_info->trans_mutex);
 822         cur = root->fs_info->running_transaction;
 823         if (!cur) {
 824                 mutex_unlock(&root->fs_info->trans_mutex);
 825                 goto out;
 826         }
 827         now = get_seconds();
 828         if (now < cur->start_time || now - cur->start_time < 30) {
 829                 mutex_unlock(&root->fs_info->trans_mutex);
 830                 delay = HZ * 5;
 831                 goto out;
 832         }
 833         mutex_unlock(&root->fs_info->trans_mutex);
 834         btrfs_defrag_dirty_roots(root->fs_info);
 835         trans = btrfs_start_transaction(root, 1);
 836         ret = btrfs_commit_transaction(trans, root);
 837 out:
 838         mutex_unlock(&root->fs_info->fs_mutex);
 839         btrfs_clean_old_snapshots(root);
 840         btrfs_transaction_queue_work(root, delay);
 841 }
 842
 843 void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
 844 {
 845         if (!root->fs_info->closing)
 846                 queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
 847 }
 848
 849 void btrfs_transaction_flush_work(struct btrfs_root *root)
 850 {
 851         cancel_delayed_work(&root->fs_info->trans_work);
 852         flush_workqueue(trans_wq);
 853 }
 854
 855 void __init btrfs_init_transaction_sys(void)
 856 {
 857         trans_wq = create_workqueue("btrfs-transaction");
 858 }
 859
 860 void btrfs_exit_transaction_sys(void)
 861 {
 862         destroy_workqueue(trans_wq);
 863 }
 864